YOLO-waste: Utilizing YOLOv4 and YOLOv4-tiny for waste classification and management

Abstract

Waste Management refers to a process that begins with the detection of various waste types and gradually manages waste from its conception through its disposal. Since it is extremely important to prevent contamination of the environment and reuse waste in as many ways as possible, this paper is believed to bring in a beneficial change in the way how waste is handled. This research used an diverse dataset for garbage classification with a lot of images. The dataset was divided into 8 different categories to train a machine learning model. A total of 16,000 photos were created by augmenting various classes, including battery, biological, cardboard, garments, green-glass, paper, plastic, and waste, to ensure equal size. This study proposes the use of YOLOv4 and YOLOv4-tiny alongside with Darknet-53 to detect waste. Leveraging YOLOv4, the model produced a promising mAP of 85.73%, precision of 0.78, recall of 0.84, F1-score of 0.81, and average IoU of 62.05%. In case of YOLOv4-tiny, mAP was 81.28%, along with precision, recall, F1-score and average IoU value of 0.60, 0.87, 0.71, and 45.67%.